Early detection, identification, and warning are essential to minimize casualties from a biological attack. For covert attacks, sick people are likely to provide the first indication of an attack. An enhanced medical surveillance system that synthesizes distributed health indicator information and rapidly analyzes the information can dramatically increase the number of lives saved. Current surveillance methods to detect both biological attacks and natural outbreaks are hindered by factors such as distributed ownership of information, incompatible data storage and analysis programs, and patient privacy concerns. Moreover, because data are not widely shared, few data mining algorithms have been tested on and applied to diverse health indicator data. This project addresses both integration of multiple data sources and development and integration of analytical tools for rapid detection of disease outbreaks. The first prototype developed included an application to query and display distributed patient records. This application incorporated need-to-know access control and incorporated data from standard commercial databases. Two different algorithms were developed and tested for outbreak recognition. The first is a pattern recognition technique that searches for space-time data clusters that may signal a disease outbreak. The second is a genetic algorithm to design and train neural networks (GANN) applied toward disease forecasting. These algorithms were tested against influenza, respiratory illness, and Dengue Fever data.
Sandia National Laboratories: http://www.sandia.gov